Generative AI and Copyright: Striking a Balance for Innovation and Protection

By Andeed Ma

Generative AI has undoubtedly ushered in a new era of innovation and convenience, enabling machines to create text, images, and various content types in response to user commands. Tools such as Chat GPT and Stable Diffusion have expanded the realm of possibilities, aiding students in their academic endeavors and inspiring imaginative artwork. However, amidst these impressive advancements lies a complex web of challenges related to copyright that necessitates careful consideration. In this article, we will explore the delicate balance between fostering innovation and protecting intellectual property rights in the context of generative AI.

Generative AI systems, including Chat GPT and Stable Diffusion, rely on vast datasets containing text and images, often sourced from copyrighted material, to train and generate creative outputs. This raises a fundamental question: should the machine learning process be subject to copyright restrictions?

Central to the generative AI and copyright debate is the process by which AI systems are trained. These systems require access to extensive datasets, which may include copyrighted works, to operate effectively. Although the specific datasets used by companies developing these AI systems remain undisclosed, there is a clear indication that the training process often involvesthe unauthorized reproduction of copyrighted content scraped from the internet. This situation raises legitimate concerns about potential copyright infringement.

For instance, Getty Images filed a lawsuit against Stability AI for allegedly copying over 12 million photographs from its collection. Artists have also taken legal action against Stability AI, Midjourney, and DeviantArt, asserting that their copyrights were infringed upon through the use of training images. These legal challenges underscore the imperative need for ethical considerations in the training of generative AI.

The output generated by generative AI systems can inadvertently infringe on copyrights. When AI produces text or images in response to user commands, it may unknowingly replicate substantial portions of copyrighted works. This is particularly evident when AI generates content closely resembling the original source, such as retaining watermarks or distinct elements.

It is essential to note, however, that copyright law does not protect artistic style. Therefore, if AI generates content inspired by a specific artist's style, it may not necessarily constitute copyright infringement.

In some jurisdictions, including Singapore, if these uses fall within either the fair use clause (section 190-191) or the computational data analysis exception (section 243-244 of the Copyright Act), Singaporean copyright law may offer a defencefor such infringing applications for AI-generated content. The "lawful access" to the copyrighted content and the fact that the making of a copy cannot be used for any other purpose other than for identifying, extracting, or analysing information/data and using that to improve the functioning of a programme in relation to that type of information/data are two of the five strict requirements that must be met in order to qualify for the computational data analysis exception. However, when AI scrapes the internet for data and uses it to generate new works, it may not meet these criteria, potentially leading to copyright infringement.

On the other hand, fair use provisions consider factors like the purpose of use and its impact on the market. In Singapore, as in the United States, transformative use is a crucial factor in determining fair use. Similar to the US fair use rule (17 USC 107), Singapore's section 191 of the Copyright Act lists a non-exclusive list of four elements to be taken into considerationwhen deciding whether an unlawful use is fair and, thus, permissible. In the US, fair use has permitted Google Books to make digital copies of tens of millions of books without the owners' consent in order to provide a publicly accessible internet search feature. Unfortunately, generative AI often falls short of demonstrating sufficient transformative purpose, as it primarily appropriates existing creative works.

Furthermore, the widespread use of generative AI without adequate compensation poses a significant threat to the licensing markets of copyrighted works, undermining the argument for fair use in many instances.

To address the complex interplay between generative AI and copyright, we can turn to the principles of risk management outlined in ISO 31000. These principles offer valuable guidance for navigating this intricate landscape while maintaining a positive perspective.

ISO 31000 emphasizes the integration of risk management into an organization's governance structure. In the context ofgenerative AI and copyright, it's crucial to integrate ethical considerations and copyright compliance into the development and deployment of AI systems. Organizations should establish clear policies and guidelines for the ethical use of AI technologies, including copyright compliance.

Organizations are encouraged to tailor risk management processes to their specific needs and context. AI developers and content creators should collaborate to customize licensing agreements and permissions that align with the unique requirements of generative AI projects. These agreements should outline the extent to which AI can use copyrighted material, ensuring compliance even before the data is fed into the model.

A structured approach to risk assessment, including risk identification, analysis, and evaluation can be conducted. In the generative AI and copyright space, this means conducting a thorough analysis of potential copyright issues during the development and training of AI models. It also involves evaluating the ethical implications of AI-generated content, including its impact on copyright holders.

Similarly, stakeholders in the generative AI ecosystem, including AI developers, content creators, and regulatory bodies, should engage in open and transparent discussions about copyright, intellectual property, and ethical considerations. This dialogue can lead to the establishment of industry best practices and standards.

ISO 31000 emphasizes the need for ongoing monitoring and review of risk management processes. In the context ofgenerative AI and copyright, this means continuously assessing the impact of AI-generated content on copyright holders and making necessary adjustments to licensing agreements and compliance measures.

Organizations to develop risk treatment plans to address identified risks. AI developers and content creators should have mechanisms in place to mitigate copyright infringement risks. This may involve implementing content filtering and recognition technologies to ensure that AI-generated content does not infringe upon copyrights.

In addition to applying ISO 31000 principles, education and awareness play a crucial role in navigating the intersection ofgenerative AI and copyright in a positive and constructive manner. Content creators should receive education about the capabilities and limitations of AI systems, enabling them to make informed decisions about their works. Likewise, AI developers must possess a comprehensive understanding of copyright laws and ethical considerations when training their models on copyrighted material.

In the end, it's about embracing the future while preserving the foundations of creative expression and innovation. A world where innovation and protection coexist is a world where everyone can benefit, and that's a future worth striving for.

About the Author

Mr. Andeed Ma is an AI cloud business and risk management leader for more than 15 years. He has contributed to major technology companies such as ServiceNow, Ivanti, ByteDance, and CyberArk. He leads a non-profit enterprise risk management association known as RIMAS (Risk and Insurance Management Association of Singapore) as their President. He lectures, speaks and mentors at the Singapore University of Social Sciences (SUSS) on Hyperautomation, Introduction to AI; and at the Singapore Management University (SMU) on the Essentials of Cloud Computing, Regulatory Technology (RegTech) and Sustainable AI. He also sits in various councils and think tanks as a contributor such as ForHumanity in UK, ISO/IEC JTC 1/SC 42 Artificial intelligence, IEC SEG 15 - Metaverse, Global Fintech Institute (GFI), Artificial Intelligence International Institute (AIII), Institute of Blockchain Singapore (IBS), and Blockchain Security Alliance. He is also a RealmIQ Mentor.